Machine-Learning

Objectives

This course focuses on supervised learning methods for regression and classification. Starting with basic algorithms such as ordinary least squares, we will cover regularization methods (essential in large-scale learning contexts), nonparametric decision rules such as support vector machines (SVMs), and more broadly kernel methods in self-reproducing kernel Hilbert spaces. Next, we will study ensemble techniques, with reminders on bagging and random forests, with a particular focus on boosting (XGBoost). Finally, we will cover some aspects of variable selection.rnThroughout the course, we will emphasize methodological and algorithmic aspects, while providing an overview of the underlying theoretical foundations. Practical sessions will allow students to apply the methods to datasets using Python. The course will alternate between lectures and practical sessions.rnrnTranslated with DeepL.com (free version)

Course outline

Supervised learning; Regression; Classification; Empirical risk minimization; Model evaluation; Cross validation; Model complexity; Large scale optimization; Stochastic gradient descent; Regularization; RIDGE and LASSO; Support Vector Machine; Kernel trick; Ensemble methods (Random forest, Aggregation, Boosting); Feature selection.

Prerequisites

Fundamentals of supervised learning, linear regression, logistic regression, fundamentals of optimization, Python programming